The field of the invention relates to the reception of multi-carrier signals. More precisely, the invention relates to estimating the phase shift introduced by demodulation operations.
The invention is applicable to all types of signals using several carrier frequencies, in other words systems using signals transmitted using the Frequency Division Multiplex (FDM) technique, for example the Coded Orthogonal Frequency Division Multiplex (COFDM) system, used particularly within the framework of the European Eureka 147 xe2x80x9cDABxe2x80x9d (Digital Audio Broadcasting) and RACE dTTb (digital Terrestrial TV broadcasting) projects.
In this type of transmission system, the source data to be transmitted are organized in symbols (composed of one or several source data) each modulating a carrier frequency chosen among several carriers, during a predetermined time interval. The signal formed by the set of modulated carriers is transmitted to one or several receivers that receive an emitted signal disturbed by the transmission channel.
In principle, demodulation usually consists of estimating the response of the transmission channel for each carrier and for each symbol, and then dividing the received signal by this estimate to obtain an estimate of the emitted symbol.
A number of demodulation techniques are known. The demodulation may be differential or coherent. For example, patent FR-94 07984 deposited by the same applicants describes one technique facilitating coherent demodulation using reference symbols (or xe2x80x9cpilotsxe2x80x9d) known in receivers and inserted regularly among the useful symbols.
Therefore, the multi-carrier digital signal considered includes a number of xe2x80x9cpilotsxe2x80x9d, in other words carriers modulated by known values of the receiver. Let C be the set of indices of these carriers. If K belongs to C, the complex value modulating carrier number k for the duration of symbol number n is denoted Pn,k. Note that this value is known to the receiver.
In reception, the value observed on the nth symbol of carrier number k is denoted Rn,k. Typically:
Rn,k=Pn,kHn,kejxcfx86n(+noise)xe2x80x83xe2x80x83(1)
where Hn,k is the complex frequency response of the channel at the frequency of carrier number k;
xcfx86n is a phase introduced by the demodulation that is to be estimated (xcfx86n is common to all carriers).
Hn,k varies slowly with time (Hn,k is not very different from Hn+1,k), whereas xcfx86n can vary considerably from one symbol to the next if the local oscillator in the receiver is badly adjusted. Therefore, estimating xcfx86n is useful to determine the frequency error of the local oscillator so that it can be corrected afterwards (Automatic Frequency Control AFC).
Furthermore, as described in the documents in French patents FR-95 10067 and FR-95 10068 deposited by the applicants who also deposited this patent application, knowledge of xcfx86n is useful to reduce the bias introduced by white frequency distortion, and take account of an estimate of this type of white distortion for demodulation.
In practice, only the variations of xcfx86n need to be estimated: the value of xcfx86o fixed by definition may be incorporated into the response of channel Hn,k. Since Hn,k is not very different from Hn+1,k, (xcfx86nxe2x88x92xcfx86nxe2x88x921), denoted xcex4n, is conventionally estimated from Rn,k using the following formula:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                                      (                                                                  R                                                  n                          ,                          k                                                                    ·                                              P                                                  n                          ,                          k                                                *                                                              )                                    ⁢                                      xe2x80x83                                    ⁢                                                                                    (                                                                              R                                                                                          n                                -                                1                                                            ,                              k                                                                                ·                                                      P                                                                                          n                                -                                1                                                            ,                              k                                                        *                                                                          )                                            ⁢                                              xe2x80x83                                                              *                                                  ]                                      }                                              (        2        )            
This known approach is explained as follows. The modulus of the square of Hn,k will be denoted An,k, and the modulus of the square of Pn,k will be denoted Qn,k. According to equation (1), Rnxe2x88x921,kP*n,k gives an estimate of Qn,kHn,kejxcfx86nxe2x88x921. Similarly, Rn,kP*nxe2x88x921,k gives an estimate of Qnxe2x88x921,kHnxe2x88x921,kejxcfx86nxe2x88x921. Finally, since Hn,k is not very different from Hnxe2x88x921,k, Hn,kH*nxe2x88x921,k is close to An,k. Therefore       δ    n    =      Arg    ⁢          {                        ∑                      k            ∈            C                          ⁢                  [                                    Q                              n                ,                k                                      ⁢                          Q                                                n                  -                  1                                ,                k                                      ⁢                          A                              n                ,                k                                      ⁢                          ⅇ                              j                ⁡                                  (                                                            φ                      n                                        -                                          φ                                              n                        -                        1                                                                              )                                                              ]                    }      
It is useful to weight the terms in the sum by Qn,k.Qnxe2x88x921,k.An,k since it reflects the reliability of each term (the phase is more reliable for larger magnitudes). Therefore, this type of estimate is particularly suitable for channels with selective frequency, for which An,k varies strongly with k.
Finally, the phase xcex4n of the sum gives an estimate of (xcfx86nxe2x88x92xcfx86nxe2x88x921) for which the reliability increases with the number of carriers of C.
In practice, values of Pn,k usually have the same modulus Q. Pn,k may be completely independent of n (continuous carriers); in this case, the sum (2) may be simplified since P*n,kPnxe2x88x921,k is always equal to Q, as follows:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                                      R                                          n                      ,                      k                                                        ·                                      R                                                                  n                        -                        1                                            ,                      k                                                        ·                                      P                                                                  n                        -                        1                                            ,                      k                                        *                                                  ]                                      }                                              (2a)            
In the following, the form of equation (2) will be maintained in order to remain general, but the form (2a) is obviously preferable in the case of continuous carriers.
In some cases, an attempt is made to estimate (xcfx86nxe2x88x92xcfx86nxe2x88x92t), where t greater than 1, rather than (xcfx86nxe2x88x92xcfx86nxe2x88x921); obviously, the method remains the same. All that is necessary is to replace subscript nxe2x88x921 by nxe2x88x92t.
The inventors have observed that although this conventional technique may be efficient under transmission conditions in which there is not much interference, it is poor in some conditions in the presence of interference sources. In this case, some carriers are continuously polluted by very strong noise, sometimes much stronger than the useful signal. The result of this is some completely distorted terms, usually with a strong amplitude, in sum (2). Since xcex4n is fairly small, particularly when the local oscillator tends towards the right frequency), these few terms can completely disturb the result of the sum.
One particular purpose of this invention is to overcome this disadvantage with the state-of-the-art.
More precisely, one purpose of the invention is to provide a technique for strongly attenuating the effect of interference sources on the estimate of the phase shift induced by the receiver.
In other words, the purpose of the invention is to provide a process and a corresponding receiver that optimizes the estimate of the phase shift, particularly when the phase shift is small.
Another purpose of the invention is to supply a corresponding technique that is simple to use and that does not require complex calculations or specific means in the receivers.
Another purpose of the invention is to provide a similar technique that does not require any modification to the signal to be emitted, and which is therefore compatible with transmission techniques already used.
These purposes, and others that will become clear in the following description, are achieved according to the invention using a process for estimating a parasite phase shift on reception of a multi-carrier signal formed by a time sequence of symbols modulating a plurality of carrier frequencies, at least some of the said carrier frequencies for at least some of the said symbols carrying reference elements, for which the values on transmission are known to the receiver that receives them, this process being of the type comprising a step in which the phase variation xcex4n between at least two symbols carrying the reference elements is estimated, in which the contribution of at least some of the said reference elements in the said estimate is weighted by information representative of the noise affecting the carrier frequency carrying the said reference element.
Thus, the importance assigned to each pilot depends on the disturbances that affect it, and therefore its credibility. It should be noted that this new approach is not obvious, but is based particularly on the formulation of the problem described above, that has never been made or envisaged before.
Note that in general, the phase shift is induced particularly (but not exclusively) by the receiver.
The said information representative of the noise may in particular belong to the group comprising:
the variance of the noise affecting each carrier frequency ("sgr"2k);
the amplitude of the product Rn,k.R*nxe2x88x92i,k;
the error rate affecting each carrier frequency;
information indicating that the carrier is not reliable.
This information, and particularly the noise variance, has often already been calculated for other applications, particularly the calculation of the signal to noise (S/N) ratio on each carrier to optimize Viterbi decoding with weighted inputs (xe2x80x9csoft decodingxe2x80x9d).
Therefore, very few calculations need to be carried out. All that is necessary is to weight normal calculations as a function of available information.
Thus, when estimating the variance, the estimate of the phase variation xcex4n is advantageously calculated as follows:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                                      1                                          σ                      k                      2                                                        ⁢                                      (                                                                  R                                                  n                          ,                          k                                                                    ·                                              P                                                  n                          ,                          k                                                *                                                              )                                    ⁢                                      xe2x80x83                                    ⁢                                                                                    (                                                                              R                                                                                          n                                -                                1                                                            ,                              k                                                                                ·                                                      P                                                                                          n                                -                                1                                                            ,                              k                                                        *                                                                          )                                            ⁢                                              xe2x80x83                                                              *                                                  ]                                      }                                              (        3        )            
where Pn,k is the value of the reference element as it was emitted;
Rn,k is the value of the reference element as it was received;
n is the time dimension;
k is the frequency dimension;
C is the set of carrier frequencies carrying reference elements.
When the receiver calculates an estimate of xcfx81n,k=An,k/"sgr"2k, where "sgr"2k is an estimate of the noise variance and An,k is an estimate of the modulus of the square of the frequency response of the channel on carrier frequency k at instant n, the estimate of the phase variation xcex4n may for example be calculated as follows:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                                                            ρ                                              n                        ,                        k                                                              ⁡                                          (                                                                        R                                                      n                            ,                            k                                                                          ·                                                  P                                                      n                            ,                            k                                                    *                                                                    )                                                        ⁢                                      xe2x80x83                                    ⁢                                                                                    (                                                                              R                                                                                          n                                -                                1                                                            ,                              k                                                                                ·                                                      P                                                                                          n                                -                                1                                                            ,                              k                                                        *                                                                          )                                            ⁢                                              xe2x80x83                                                              *                                                  ]                                      }                                              (        4        )            
According to one advantageous embodiment of the invention, the said weighting consists simply of thresholding, and only reference elements carried by carrier frequencies for which the said information representative of the noise exceeds a predetermined threshold are taken into account.
It is verified that this solution, which is shorter but simpler, is sufficient in most cases.
In this case, if the said information representative of the noise is an estimate of the noise variance, the estimate of the phase variation xcex4n may be calculated as follows:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                                      I                    ⁡                                          (                                              σ                        k                                            )                                                        ⁢                                      (                                                                  R                                                  n                          ,                          k                                                                    ·                                              P                                                  n                          ,                          k                                                *                                                              )                                    ⁢                                      xe2x80x83                                    ⁢                                                                                    (                                                                              R                                                                                          n                                -                                1                                                            ,                              k                                                                                ·                                                      P                                                                                          n                                -                                1                                                            ,                              k                                                        *                                                                          )                                            ⁢                                              xe2x80x83                                                              *                                                  ]                                      }                                              (        5        )            
where: I("sgr"k) is equal to 1 provided that "sgr"k is below the said threshold, and is otherwise equal to 0.
If the said information representative of the noise is an estimate of xcfx81n,k=An,k/"sgr"2k, the estimate of the phase variation xcex4n is advantageously calculated as follows:                               δ          n                =                  Arg          ⁢                      {                                          ∑                                  k                  ∈                  C                                            ⁢                              [                                  J                  ⁢                                      xe2x80x83                                    ⁢                                      (                                          ρ                                              n                        ,                        k                                                              )                                    ⁢                                      (                                                                  R                                                  n                          ,                          k                                                                    ·                                              P                                                  n                          ,                          k                                                *                                                              )                                    ⁢                                      xe2x80x83                                    ⁢                                                                                    (                                                                              R                                                                                          n                                -                                1                                                            ,                              k                                                                                ·                                                      P                                                                                          n                                -                                1                                                            ,                              k                                                        *                                                                          )                                            ⁢                                              xe2x80x83                                                              *                                                  ]                                      }                                              (        6        )            
where J(xcfx81n,k) is equal to 1 provided that xcfx81k is above the said threshold, and is otherwise equal to 0.
The invention also relates to receivers embodying this type of process. These receivers comprise estimating means that weight the contribution of each of the said reference elements in the said estimate by information representative of the noise affecting the carrier frequency carrying the said reference element.
Preferably, the said information representative of the noise is an estimate of the noise variance affecting each carrier frequency ("sgr"2k).